pith. sign in

arxiv: 2605.04806 · v1 · submitted 2026-05-06 · 💻 cs.RO

Dr-PoGO: Direct Radar Pose-Graph Optimization

Cedric Le Gentil , Weican Li , Leonardo Brizi , Timothy D. Barfoot This is my paper

Pith reviewed 2026-05-08 17:12 UTC · model grok-4.3

classification 💻 cs.RO
keywords radar SLAMdirect registrationpose graph optimizationodometryloop closureadverse weather perception
0
0 comments X

The pith

A direct radar pose-graph optimization method delivers state-of-the-art SLAM performance over 300 km of real-world data.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces Dr-PoGO as a SLAM method for 2D spinning radar that performs direct registration of radar data for both odometry and loop closures. It uses an off-the-shelf place recognition to find loops and a coarse-to-fine approach with visual features to initialize the direct radar transform refinement. These are then used in a global pose-graph optimization to build the map and trajectory. This approach is designed to work reliably in poor visibility conditions like dust, snow, and rain where other sensors struggle. It reports superior results compared to existing methods on over 300 kilometers of real automotive data.

Core claim

Dr-PoGO leverages direct registration techniques for radar odometry and loop-closure registration, initialized via visual features for the latter, and optimizes the global trajectory in a pose-graph to enable robust SLAM in adverse weather.

What carries the argument

Direct radar registration combined with coarse-to-fine loop closure using visual initial guesses, embedded in pose-graph optimization.

If this is right

  • Provides ego-motion estimation without needing to extract point clouds or features from radar scans.
  • Allows loop closures even when place recognition does not output transformations.
  • Supports reliable mapping and localization regardless of weather conditions.
  • Demonstrates improved performance on long-term real-world automotive datasets.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method may allow radar to serve as a primary sensor in SLAM systems for all-weather autonomy.
  • Future work could explore integrating this direct approach with other sensor types for multi-modal fusion.
  • Testing on datasets with more severe occlusions could further validate the convergence of the direct refinement step.

Load-bearing premise

The visual feature-based initial guess for loop closures is sufficiently close to the true transformation for the direct radar registration to converge to the global optimum.

What would settle it

Observing a case where the radar direct registration converges to an incorrect alignment despite the visual initial guess, leading to inconsistent loop closures and trajectory errors.

Figures

Figures reproduced from arXiv: 2605.04806 by Cedric Le Gentil, Leonardo Brizi, Timothy D. Barfoot, Weican Li.

Figure 1
Figure 1. Figure 1: Dr-PoGO performs SLAM using 2D radar data. It is validated view at source ↗
Figure 2
Figure 2. Figure 2: Block-diagram overview of Dr-PoGO. First, DRO processes the raw radar data (and optional yaw-gyroscope measurements) to generate per-scan view at source ↗
Figure 3
Figure 3. Figure 3: Example of local-map registration. Alone, the cross-correlation view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of the challenging conditions in Boreas-RT view at source ↗
Figure 5
Figure 5. Figure 5: Absolute Trajectory Error and End-Pose Error (log scale) for each view at source ↗
Figure 6
Figure 6. Figure 6: Trajectory examples over the three environment types of the self view at source ↗
Figure 7
Figure 7. Figure 7: Strip plots of the position (left column) and rotation (right column) view at source ↗
read the original abstract

This paper introduces Dr-PoGO, a method for Simultaneous Localization And Mapping (SLAM) using a 2D spinning radar. Unlike cameras or lidars that require line-of-sight, millimetre-wave radars can `see' through dust, falling snow, rain, etc. Accordingly, it is a great modality for robust perception regardless of the weather conditions. While most existing radar-based SLAM methods rely on the extraction of point clouds or features to perform ego-motion estimation, Dr-PoGO leverages direct registration techniques for odometry (DRO) and loop-closure registration. An off-the-shelf radar-focused place recognition algorithm, RaPlace, provides loop-closure candidates. As RaPlace does not provide relative transformations, Dr-PoGO introduces a coarse-to-fine registration that uses visual features and descriptors to obtain an initial guess for the direct transformation refinement. The global trajectory is optimized in a pose-graph optimization. Dr-PoGO demonstrates state-of-the-art performance over 300km of data in various real-world automotive environments. Our implementation is publicly available: https://github.com/utiasASRL/dr_pogo.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 2 minor

Summary. The paper introduces Dr-PoGO, a radar SLAM system for 2D spinning radar that performs direct radar odometry (DRO), obtains loop-closure candidates via the off-the-shelf RaPlace place-recognition algorithm, applies a coarse-to-fine registration that first uses visual features and descriptors to seed an initial guess before direct radar refinement, and finally optimizes the trajectory in a pose-graph framework. It reports state-of-the-art performance across 300 km of real-world automotive sequences in varied environments and releases the implementation publicly.

Significance. If the empirical results hold, the work advances radar-based SLAM by showing that direct registration can be applied successfully to both odometry and loop closure without intermediate feature extraction, exploiting radar's robustness to weather and occlusion. The scale of the evaluation (300 km) and the public code release are concrete strengths that support reproducibility and further research in adverse-condition perception.

minor comments (2)
  1. Abstract: the SOTA claim would be more informative if it named the primary competing methods and the quantitative metrics (e.g., absolute trajectory error) on which superiority is asserted.
  2. The coarse-to-fine registration paragraph does not specify which visual feature detector and descriptor are employed, limiting immediate reproducibility of the initial-guess stage.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive assessment of Dr-PoGO, including the recognition of our 300 km evaluation and public code release. The recommendation for minor revision is noted, and we will incorporate any editorial improvements in the revised version.

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper describes an engineering pipeline that chains external modules (RaPlace place recognition, direct radar odometry/refinement, visual-feature coarse alignment, and standard pose-graph optimization) without reducing any claimed performance metric or transformation to a fitted parameter or self-referential equation. The 300 km SOTA result is presented as an empirical outcome of the integrated system rather than a mathematical consequence of its own definitions. Any self-citations for prior direct-registration components are not load-bearing for the uniqueness or correctness of the central claims.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach rests on domain assumptions about the applicability of direct registration to radar scans and the reliability of RaPlace candidates; no new entities are postulated and no numerical parameters are fitted inside the abstract description.

axioms (2)
  • domain assumption Direct registration techniques developed for radar data can produce accurate ego-motion estimates without explicit feature extraction.
    Invoked for both DRO and loop-closure refinement.
  • domain assumption RaPlace supplies loop-closure candidates of sufficient quality that the subsequent coarse-to-fine registration succeeds.
    Used to trigger the registration pipeline.

pith-pipeline@v0.9.0 · 5503 in / 1368 out tokens · 23743 ms · 2026-05-08T17:12:02.383768+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages

  1. [1]

    A solution to the simultaneous localization and map building (slam) problem,

    M. Dissanayake, P. Newman, S. Clark, H. Durrant-Whyte, and M. Csorba, “A solution to the simultaneous localization and map building (slam) problem,”IEEE Trans. on Robotics and Automation, vol. 17, no. 3, pp. 229–241, 2001

    work page 2001

  2. [2]

    A new wave in robotics: Survey on recent mmwave radar applications in robotics,

    K. Harlow, H. Jang, T. D. Barfoot, A. Kim, and C. Heckman, “A new wave in robotics: Survey on recent mmwave radar applications in robotics,”IEEE Trans. on Robotics (TRO), vol. 40, pp. 4544–4560, 2024

    work page 2024

  3. [3]

    Do we still need to work on odometry for autonomous driving?

    C. Le Gentil, D. Lisus, and T. D. Barfoot, “Do we still need to work on odometry for autonomous driving?” inICRA 2025 Workshop on Field Robotics, 2025

    work page 2025

  4. [4]

    Dro: Doppler-aware direct radar odometry,

    C. Le Gentil, L. Brizi, D. Lisus, X. Qiao, G. Grisetti, and T. D. Barfoot, “Dro: Doppler-aware direct radar odometry,” inProc. of Robotics: Science and Systems (RSS), 2025

    work page 2025

  5. [5]

    Raplace: Place recognition for imaging radar using radon transform and mutable threshold,

    H. Jang, M. Jung, and A. Kim, “Raplace: Place recognition for imaging radar using radon transform and mutable threshold,” inProc. of the IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS), 2023, pp. 11 194–11 201

    work page 2023

  6. [6]

    Radar odometry for au- tonomous ground vehicles: A survey of methods and datasets,

    N. J. Abu-Alrub and N. A. Rawashdeh, “Radar odometry for au- tonomous ground vehicles: A survey of methods and datasets,”IEEE Trans. on Intelligent Vehicles (T-IV), vol. 9, no. 3, pp. 4275–4291, 2024

    work page 2024

  7. [7]

    Millimeter wave fmcw radars for perception, recognition and localization in automotive applications: A survey,

    A. Venon, Y . Dupuis, P. Vasseur, and P. Merriaux, “Millimeter wave fmcw radars for perception, recognition and localization in automotive applications: A survey,”IEEE Trans. on Intelligent Vehicles (T-IV), vol. 7, pp. 1–1, 09 2022

    work page 2022

  8. [8]

    Distinctive image features from scale-invariant key- points,

    D. G. Lowe, “Distinctive image features from scale-invariant key- points,”Intl. Journal of Computer Vision (IJCV), vol. 60, no. 2, pp. 91–110, Nov. 2004

    work page 2004

  9. [9]

    Radar slam using visual features,

    J. Callmer, D. T ¨ornqvist, F. Gustafsson, H. Svensson, and P. Carlbom, “Radar slam using visual features,”EURASIP Journal on Advances in Signal Processing, vol. 2011, 12 2011

    work page 2011

  10. [10]

    Radar cfar thresholding in clutter and multiple target situations,

    H. Rohling, “Radar cfar thresholding in clutter and multiple target situations,”IEEE Trans. on Aerospace and Electronic Systems (T-AES), vol. AES-19, no. 4, pp. 608–621, 1983

    work page 1983

  11. [11]

    Cfear radarodometry - conservative filtering for efficient and accurate radar odometry,

    D. Adolfsson, M. Magnusson, A. Alhashimi, A. Lilienthal, and H. An- dreasson, “Cfear radarodometry - conservative filtering for efficient and accurate radar odometry,” 09 2021, pp. 5462–5469

    work page 2021

  12. [12]

    Lidar-level localization with radar? the cfear ap- proach to accurate, fast, and robust large-scale radar odometry in diverse environments,

    D. Adolfsson, M. Magnusson, A. Alhashimi, A. J. Lilienthal, and H. Andreasson, “Lidar-level localization with radar? the cfear ap- proach to accurate, fast, and robust large-scale radar odometry in diverse environments,”IEEE Trans. on Robotics (TRO), vol. 39, no. 2, pp. 1476–1495, 2023

    work page 2023

  13. [13]

    The Finer Points: A Systematic Comparison of Point-Cloud Extractors for Radar Odom- etry,

    E. Preston-Krebs, D. Lisus, and T. D. Barfoot, “The Finer Points: A Systematic Comparison of Point-Cloud Extractors for Radar Odom- etry,”Proceedings of the Conference on Robots and Vision, may 27 2025

    work page 2025

  14. [14]

    Do we need to compensate for motion distortion and doppler effects in spinning radar navigation?

    K. Burnett, A. P. Schoellig, and T. D. Barfoot, “Do we need to compensate for motion distortion and doppler effects in spinning radar navigation?”IEEE Robotics and Automation Letters (RA-L), vol. 6, no. 2, pp. 771–778, 2021

    work page 2021

  15. [15]

    Are We Ready for Radar to Replace Lidar in All-weather Mapping and Localization?

    K. Burnett, Y . Wu, D. J. Yoon, A. P. Schoellig, and T. D. Barfoot, “Are We Ready for Radar to Replace Lidar in All-weather Mapping and Localization?”IEEE Robotics and Automation Letters (RA-L), vol. 7, no. 4, pp. 10 328–10 335, 2022

    work page 2022

  16. [16]

    Continuous-time radar- inertial and lidar-inertial odometry using a gaussian process motion prior,

    K. Burnett, A. P. Schoellig, and T. D. Barfoot, “Continuous-time radar- inertial and lidar-inertial odometry using a gaussian process motion prior,”IEEE Trans. on Robotics (TRO), vol. 41, p. 1059–1076, Jan. 2025

    work page 2025

  17. [17]

    Radar Odometry Combining Probabilistic Estimation and Unsupervised Fea- ture Learning,

    K. Burnett, D. J. Yoon, A. P. Schoellig, and T. D. Barfoot, “Radar Odometry Combining Probabilistic Estimation and Unsupervised Fea- ture Learning,” inProc. of Robotics: Science and Systems (RSS), 2021

    work page 2021

  18. [18]

    Pointing the Way: Refining Radar- Lidar Localization Using Learned ICP Weights,

    Lisus, Daniil and Laconte, Johann and Burnett, Keenan and Zhang, Ziyu and Barfoot, Timothy D., “Pointing the Way: Refining Radar- Lidar Localization Using Learned ICP Weights,”Proceedings of the Conference on Robots and Vision, may 27 2025

    work page 2025

  19. [19]

    Radar scan matching slam using the fourier-mellin transform,

    P. Checchin, F. G ´erossier, C. Blanc, R. Chapuis, and L. Trassoudaine, “Radar scan matching slam using the fourier-mellin transform,” vol. 62, 01 2009, pp. 151–161

    work page 2009

  20. [20]

    Masking by Moving: Learning Distraction-Free Radar Odometry from Pose Information,

    D. Barnes, R. Weston, and I. Posner, “Masking by Moving: Learning Distraction-Free Radar Odometry from Pose Information,” inConf. Robot Learn., 2019

    work page 2019

  21. [21]

    Pharao: Direct radar odometry using phase correlation,

    Y . S. Park, Y .-S. Shin, and A. Kim, “Pharao: Direct radar odometry using phase correlation,” inProc. of the IEEE Intl. Conf. on Robotics & Automation (ICRA), 2020, pp. 2617–2623

    work page 2020

  22. [22]

    Are doppler velocity measurements useful for spinning radar odometry?

    D. Lisus, K. Burnett, D. J. Yoon, R. Poulton, J. Marshall, and T. D. Barfoot, “Are doppler velocity measurements useful for spinning radar odometry?”IEEE Robotics and Automation Letters (RA-L), vol. 10, no. 1, pp. 224–231, 2025

    work page 2025

  23. [23]

    Real-time pose graph slam based on radar,

    M. Holder, S. Hellwig, and H. Winner, “Real-time pose graph slam based on radar,” in2019 IEEE Intelligent Vehicles Symposium (IV), 2019, pp. 1145–1151

    work page 2019

  24. [24]

    Radarslam: A robust simultaneous localization and mapping system for all weather conditions,

    Z. Hong, Y . Petillot, A. Wallace, and S. Wang, “Radarslam: A robust simultaneous localization and mapping system for all weather conditions,”Intl. Journal of Robotics Research (IJRR), vol. 41, no. 5, pp. 519–542, 2022

    work page 2022

  25. [25]

    Tbv radar slam – trust but verify loop candidates,

    D. Adolfsson, M. Karlsson, V . Kubelka, M. Magnusson, and H. An- dreasson, “Tbv radar slam – trust but verify loop candidates,”IEEE Robotics and Automation Letters (RA-L), vol. 8, no. 6, pp. 3613–3620, 2023

    work page 2023

  26. [26]

    Large scale place recognition in 2d lidar scans using geometrical landmark relations,

    M. Himstedt, J. Frost, S. Hellbach, H.-J. B ¨ohme, and E. Maehle, “Large scale place recognition in 2d lidar scans using geometrical landmark relations,” inProc. of the IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS), 2014, pp. 5030–5035

    work page 2014

  27. [27]

    M2dp: A novel 3d point cloud descriptor and its application in loop closure detection,

    L. He, X. Wang, and H. Zhang, “M2dp: A novel 3d point cloud descriptor and its application in loop closure detection,” inProc. of the IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS), 2016, pp. 231–237

    work page 2016

  28. [28]

    Scan context++: Structural place recog- nition robust to rotation and lateral variations in urban environments,

    G. Kim, S. Choi, and A. Kim, “Scan context++: Structural place recog- nition robust to rotation and lateral variations in urban environments,” IEEE Trans. on Robotics (TRO), vol. 38, no. 3, pp. 1856–1874, 2022

    work page 2022

  29. [29]

    Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,

    M. A. Fischler and R. C. Bolles, “Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,”Communications of the ACM, vol. 24, no. 6, pp. 381–395, 1981

    work page 1981

  30. [30]

    Boreas: A multi-season autonomous driving dataset,

    K. Burnett, D. J. Yoon, Y . Wu, A. Z. Li, H. Zhang, S. Lu, J. Qian, W.- K. Tseng, A. Lambert, K. Y . Leung, A. P. Schoellig, and T. D. Barfoot, “Boreas: A multi-season autonomous driving dataset,”Intl. Journal of Robotics Research (IJRR), vol. 42, no. 1-2, pp. 33–42, 2023

    work page 2023

  31. [31]

    Boreas road trip: A multi-sensor autonomous driving dataset on challenging roads,

    D. Lisus, K. M. Papais, C. Le Gentil, E. Preston-Krebs, A. Lambert, K. Y . K. Leung, and T. D. Barfoot, “Boreas road trip: A multi-sensor autonomous driving dataset on challenging roads,” 2026. [Online]. Available: https://arxiv.org/abs/2602.16870

    work page arXiv 2026

  32. [32]

    Github - gisbi-kim/navtech-radar-slam: Real-time radar slam: Orora + scancontext

    Gisbi-Kim, “Github - gisbi-kim/navtech-radar-slam: Real-time radar slam: Orora + scancontext.” [Online]. Available: https://github.com/ gisbi-kim/navtech-radar-slam

    work page

  33. [33]

    Orora: Outlier-robust radar odometry,

    H. Lim, K. Han, G. Shin, G. Kim, S. Hong, and H. Myung, “Orora: Outlier-robust radar odometry,” inProc. of the IEEE Intl. Conf. on Robotics & Automation (ICRA), 2023, pp. 2046–2053

    work page 2023

  34. [34]

    Scan context: Egocentric spatial descriptor for place recognition within 3D point cloud map,

    G. Kim and A. Kim, “Scan context: Egocentric spatial descriptor for place recognition within 3D point cloud map,” inProc. of the IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS), Madrid, Oct. 2018

    work page 2018

  35. [35]

    Coral: Introspection for robust radar and lidar perception in diverse environments using differential entropy,

    D. Adolfsson, M. Castellano-Quero, M. Magnusson, A. J. Lilienthal, and H. Andreasson, “Coral: Introspection for robust radar and lidar perception in diverse environments using differential entropy,”Journal on Robotics and Autonomous Systems (RAS), vol. 155, p. 104136, 2022

    work page 2022

  36. [36]

    Fast-lio2: Fast direct lidar- inertial odometry,

    W. Xu, Y . Cai, D. He, J. Lin, and F. Zhang, “Fast-lio2: Fast direct lidar- inertial odometry,”IEEE Trans. on Robotics (TRO), vol. 38, no. 4, pp. 2053–2073, 2022

    work page 2053

  37. [37]

    2fast-2lamaa: Large-scale lidar-inertial localization and mapping with continuous distance fields,

    C. Le Gentil, R. Falque, D. Lisus, and T. D. Barfoot, “2fast-2lamaa: Large-scale lidar-inertial localization and mapping with continuous distance fields,” 2025. [Online]. Available: https://arxiv.org/abs/2410. 05433

    work page 2025